Using Primitives in Learning From Observation
Using Primitives in Learning From Observation
复制标题
使用原语从观察中学习
DOI：
--
复制
发表时间：
2000
期刊：
影响因子：
--
通讯作者：
C. Atkeson
中科院分区：
文献类型：
--
作者：
D. Bentivegna;C. Atkeson
本文描述了运动原语在机器人观察学习中的使用。数据是在人类执行任务时收集的。该数据被解析为称为原语的任务的小部分。为每个基元创建模块，对基元执行期间所需的运动进行编码。另一个模块对执行低级原语的时间和地点进行编码。然后，学习代理使用这些模块来执行任务。目前正在使用一个代理来测试该方法的可行性，该代理正在被教导玩虚拟和硬件弹珠迷宫游戏以及虚拟空气曲棍球游戏。它还正在测试教人形机器人与人类玩家打空气曲棍球。
This paper describes the use of motor primitives in robot learning from observation. Data is collected while a human performs a task. This data is parsed into small parts of the task called primitives. Modules are created for each primitive that encode the movements required during the performance of the primitive. Another module encodes when and where the low-level primitives are performed. A learning agent then uses these modules to perform the task. The feasibility of this method is currently being tested with an agent that is being taught to play a virtual and a hardware marble maze game and a virtual air hockey game. It is also being tested in teaching a humanoid robot to play air hockey against a human player.